1 paper across 1 session
We find that rerouting spurious shortcuts in adapter training enables robust disentanglement for text-to-image generation with adapters.
